Although displacement measurement is essential for many civil infrastructure applications, the precise estimation of structural displacement remains a challenge. In this study, a structural displacement estimation technique was developed by fusing asynchronous acceleration and computer vision measurements using a Kalman filter. First, the scale factor, which converts translation from vision measurements (in pixel units) into displacement (in length units), is automatically computed using a natural target (i.e., without any artificial target or any prior knowledge of the target size). Second, an improved feature matching algorithm was developed to better trace the natural target within the computer vision. Third, an adaptive multirate Kalman filter was formulated such that asynchronous computer vision and acceleration measurements with different sampling rates could be seamlessly combined to improve displacement estimation. The feasibility and effectiveness of the proposed displacement estimation technique were validated by performing shaking table, four-story building model, and steel box girder pedestrian bridge tests. In all tests, the proposed technique was able to accurately estimate displacements with root mean square errors of less than 3 mm.

中文翻译：

---

通过融合异步加速度和计算机视觉测量的实时结构位移估计

尽管位移测量对于许多民用基础设施应用来说是必不可少的，但结构位移的精确估计仍然是一个挑战。在这项研究中，通过使用卡尔曼滤波器融合异步加速度和计算机视觉测量，开发了一种结构位移估计技术。首先，将视觉测量（以像素单位）转换为位移（以长度单位）的比例因子是使用自然目标自动计算的（即，没有任何人工目标或目标大小的任何先验知识）。其次，开发了一种改进的特征匹配算法，以更好地跟踪计算机视觉中的自然目标。第三，制定了自适应多速率卡尔曼滤波器，以便可以无缝组合具有不同采样率的异步计算机视觉和加速度测量，以改进位移估计。通过振动台、四层建筑模型和钢箱梁人行桥试验验证了所提出的位移估计技术的可行性和有效性。在所有测试中，所提出的技术能够以小于 3 mm 的均方根误差准确估计位移。